Hi everyone!
We’re MEV-X, a research-driven project that has been actively working on MEV analysis, extraction, and solving. We’ve spent a lot of time studying how MEV strategies work in practice, and we use that experience to explore their underlying mechanics through research.
Sealed-bid auctions are widely used across blockchain systems, especially in contexts where multiple agents compete for the same opportunity, like atomic arbitrage. These auctions can occur at high frequency and are almost entirely automated. Still, the behavior of participants in these settings remains poorly characterized. There is currently no systematic understanding of how strategies change over time, how agents react to wins or losses, or what kinds of patterns form across repeated rounds.
We’ve recently started a focused research effort on this topic and plan to publish a series of articles exploring how agents behave in repeated sealed-bid auctions. Our first paper focuses on sealed-bid auctions in Polygon’s Atlas infrastructure, specifically the FastLane MEV relay. We analyzed bidding behavior in atomic arbitrage and trained reinforcement learning agents to operate under realistic latency and observability constraints. The results show that adaptive strategies emerge naturally over time, and that naive bidding often leads to value destruction. Full paper available here: [2510.14642] The Bidding Games: Reinforcement Learning for MEV Extraction on Polygon Blockchain
Although this study focused on the Atlas/FastLane setting, the modeling approach and agent framework are general and can be adapted to other sealed-bid auction formats across networks.
The recent Flashbots request for community research on “Documenting Repeated Game MEV Bidding Dynamics in Sealed Bid Auction” strongly resonated with our current line of work, it raises exactly the kind of questions we’ve been curious about and trying to understand more deeply.The fact that this is still an open area with little formal documentation makes it a great fit for our current research focus. We’re now preparing a grant application and would appreciate any guidance on how to align it with Flashbots priorities. We’d also be happy to collaborate more closely - whether through informal mentorship, joint scoping of the research questions, or other forms of cooperation that could help make the work more useful to the ecosystem.
Appreciate any input from folks thinking about similar problems!